Operation monitoring system based on flexible manufacturing job shop multi-agv system

Abstract

The application relates to the technical field of monitoring management, and discloses an operation monitoring and supervising system based on a multi-AGV system in a flexible manufacturing job shop, which comprises the following steps: collecting point cloud data, voxel gridding the point cloud data to obtain a voxel set, and determining the point cloud density of each voxel in the voxel set; calculating the global average point cloud density of the voxel set, screening low-density voxels in the voxel set based on the global average point cloud density; performing connected domain analysis on the low-density voxels in the marked voxel set, forming M low-density connected domains; performing upper and lower cone fitting processing on the mth low-density connected domain to determine a fitting error, and marking the cavity of the mth low-density connected domain based on the fitting error; determining the boundary average point cloud density of the cavity of the mth low-density connected domain, and performing point cloud filling processing on the cavity of the mth low-density connected domain according to the boundary average point cloud density to form an updated cavity of the mth low-density connected domain.

Description

Technical Field

[0001] This invention relates to the field of monitoring and management technology, and more specifically, to an operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop. Background Technology

[0002] With the increasing prevalence of flexible manufacturing workshops, multi-AGV systems have become a key component of logistics automation and collaborative manufacturing. However, to achieve high-density dynamic path scheduling and accurate map positioning, AGV systems heavily rely on LiDAR to acquire 3D environmental point cloud data for mapping and obstacle avoidance analysis.

[0003] In practical applications, high concentrations of particulate matter (such as metal fumes, welding powder, and oil mist) are often present in environments such as welding units, oil spraying and cooling zones, and cutting fluid splashing areas. These particles easily cause refraction, reflection, or direct attenuation of laser signals, resulting in sparse, fragmented, or even completely hollow cavities in local areas of the point cloud data. In point cloud representation, these cavities manifest as invalid echoes, spatially resembling a funnel-shaped hollow. Their symmetrical, cone-shaped structure, abrupt changes in boundary density, and lack of internal echo points severely interfere with map construction and path planning.

[0004] Traditional point cloud mapping and filling methods struggle to accurately identify and reconstruct the aforementioned cavity phenomena with strong spatial geometric characteristics, leading to the following technical problems: The AGV may misjudge the funnel-shaped hollow area as an obstacle, causing it to detour or stop, or misjudge the funnel-shaped hollow area as an empty area, causing it to enter an empty position. In summary, traditional point cloud mapping and filling methods lack the ability to identify and quantify the risk level of such abnormal structural regions, which prevents the scheduling system from dynamically adjusting its operating strategy based on environmental confidence information. Summary of the Invention

[0005] This invention provides an operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop, which solves the technical problems mentioned in the background art.

[0006] This invention provides an operation monitoring system for multiple AGV systems within a flexible manufacturing workshop, comprising: The data acquisition module is used to acquire point cloud data, perform voxel rasterization on the point cloud data to obtain a voxel set, and determine the point cloud density of each voxel in the voxel set. The voxel filtering module is used to calculate the global average point cloud density of the voxel set, and to filter low-density voxels in the voxel set based on the global average point cloud density; and to perform connected component analysis on the low-density voxels in the marked voxel set to form M low-density connected components. The cavity identification module is used to perform upper and lower cone fitting processing on the m-th low-density connected domain to determine the fitting error, and to label the cavity of the m-th low-density connected domain based on the fitting error. The point cloud filling module is used to determine the boundary average point cloud density of the cavity of the m-th low-density connected domain, and to perform point cloud filling processing on the cavity of the m-th low-density connected domain based on the boundary average point cloud density to form the updated cavity of the m-th low-density connected domain. The management operation module is used to determine the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain, and to trigger the management operation of the AGV system based on the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain; wherein, the management operation includes: rerouting management operation or deceleration management operation.

[0007] Further, point cloud data is collected, and voxel rasterization is performed on the point cloud data to obtain a voxel set. The point cloud density of each voxel in the voxel set is determined, including: The point cloud data is divided using resolution r to generate a set of voxels; The point cloud density corresponding to each voxel in the voxel set is the sum of the number of points within the voxel. The ratio of .

[0008] Furthermore, the global average point cloud density of the voxel set is calculated to filter low-density voxels in the voxel set based on the global average point cloud density. Connectivity analysis is then performed on the low-density voxels in the labeled voxel set to form M low-density connected components, including: The global average point cloud density of the voxel set is calculated, and the product of the preset scaling factor and the global average point cloud density is used as the filtering threshold. Voxels with point cloud density less than the screening threshold in the voxel set are classified as low-density voxels. The 26 neighborhoods of low-density voxels in the voxel set are merged to form M low-density connected regions.

[0009] Furthermore, the m-th low-density connected component undergoes upper and lower cone fitting to determine the fitting error, including: Extract the boundary point cloud set of the m-th low-density connected component; The line connecting the data acquisition module and the target suspended object's origin is used as the observation axis. The axial projection value of the line connecting each boundary point cloud in the boundary point cloud set to the target suspended object's origin is determined in the observation axis direction. Based on the axial projection value of each boundary point cloud, the boundary point cloud set is divided into: a lower region set, a transition region set, and an upper region set. The lower centroid point cloud, transition centroid point cloud, and upper centroid point cloud of the coverage area of ​​the lower region set, transition region set, and upper region set are determined respectively. With the transition centroid point cloud as the constraint center, a symmetry axis is fitted. The symmetry axis is obtained by minimizing the radial deviation distance between the lower centroid point cloud and the upper centroid point cloud and the symmetry axis. For the boundary point clouds in each set of the lower region set and the upper region set, the projection points of the boundary point clouds on the symmetry axis are determined, and the projection points, boundary point clouds, and transition centroid point clouds are connected in sequence to form a triangle. The included angle between the two sides adjacent to the transition centroid point cloud in the triangle is determined. If the included angle is within a preset angle range, the corresponding boundary point cloud is marked as a candidate boundary point cloud. N sectional planes are loaded with a symmetry axis. If the ratio of the number of candidate boundary point clouds on both sides of the nth sectional plane is not within the preset ratio range, the upper and lower limits of the preset angle range are scaled down proportionally to regenerate the candidate boundary point cloud until the ratio of the number of candidate boundary point clouds on both sides of the N sectional planes is within the preset ratio range.

[0010] Furthermore, the m-th low-density connected component undergoes upper and lower cone fitting to determine the fitting error, which also includes: K-clustering is performed on the candidate boundary point cloud to form a lower funnel sub-cluster set and an upper funnel sub-cluster set. The lower funnel sub-cluster set and the upper funnel sub-cluster set are fitted with random sampling consensus algorithm to obtain the lower cone model and the upper cone model respectively. The lower cone model includes: the lower cone vertex, the lower axis unit vector, and the lower half-vertex angle; the upper cone model includes: the upper cone vertex, the upper axis unit vector, and the upper half-vertex angle. The fitting error is constructed based on the difference between the lower half-vertex angle and the upper half-vertex angle, the sum of the lower axis unit vector and the upper axis unit vector, and the vertical distance difference between the lower cone vertex and the upper cone vertex.

[0011] Furthermore, based on the fitting error, the cavity of the m-th low-density connected region is labeled, including: Calculate the ratio of the fitting error to the preset cavity threshold; If the ratio is greater than 1, then the m-th low-density connected component is marked as a cavity; If the ratio is ≤1, no marking is performed.

[0012] Further, the boundary average point cloud density of the cavity of the m-th low-density connected domain is determined, and point cloud filling processing is performed on the cavity of the m-th low-density connected domain based on the boundary average point cloud density to form the updated cavity of the m-th low-density connected domain, including: Identify the voxels in the m-th low-density connected region that are in contact with the lower cone model or the upper cone model, form a boundary voxel set, and calculate the boundary average point cloud density of the boundary voxel set; The horizontal reference height is obtained by taking the difference between the horizontal heights of the lower cone vertex and the upper cone vertex of the lower cone model or the upper cone model, and then taking half of the difference as the horizontal height of the lower cone vertex. Construct a horizontal target density function based on the horizontal reference height and the boundary average point cloud density; For the horizontal target density function, determine the target point cloud density of each voxel in the m-th low-density connected region; based on the point cloud density difference between the target point cloud density and the point cloud density of each voxel in the m-th low-density connected region, determine the filling quantity of the virtual point cloud of the corresponding voxel, and perform uniform filling processing based on the filling quantity of the virtual point cloud to form the updated cavity of the m-th low-density connected region.

[0013] Furthermore, the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected region are determined, including: Determine the number of virtual point clouds and the point cloud density in the vth voxel of the boundary voxel set, and calculate the voxel confidence and voxel occupancy probability of the vth voxel, as follows: in, This represents the voxel confidence level of the vth voxel. Let v represent the voxel occupancy probability, specifically the voxel occupancy probability of the vth voxel. This represents the confidence decay coefficient. This represents the number of virtual point clouds filled for the v-th v element. Represents the natural base. This represents the cumulative occupancy coefficient. This represents the point cloud density of the vth voxel.

[0014] Furthermore, based on the voxel confidence and voxel occupancy probability of each voxel in the updated cavity of the m-th low-density connected component, the management operation of the AGV system is triggered, including: Determine the average voxel confidence of the boundary voxel set and average voxel occupancy probability ; Calculate the cavity risk score based on the average voxel confidence level and the average voxel occupancy probability: in, Indicates the cavity risk score; The cavity risk score is compared with a preset risk threshold. When the cavity risk score is greater than or equal to the preset risk threshold, the AGV system's management operation is triggered, as follows: If it is determined that the number of switching paths of the device to which the AGV system belongs is ≥2 at the trigger time, then the device to which the AGV system belongs will be controlled to perform a rerouting management operation. If the number of switching paths in the device to which the AGV system belongs at the trigger time is 1, then the device to which the AGV system belongs will be controlled to perform deceleration management operations. When no low-density connected domain in the voxel set is marked as a cavity, the management operation of the AGV system is not triggered.

[0015] The beneficial effects of this invention are as follows: by utilizing the cavity phenomenon of suspended objects in the point cloud spatial structure, a complete operation and supervision process is established, from the identification of low-density connected components in the point cloud, fitting of upper and lower cone models, point cloud filling driven by boundary density, and joint risk scoring of voxel confidence and occupancy probability. This enables intelligent identification, accurate reconstruction, and dynamic control of abnormal cavities in the AGV operating space, significantly improving the map stability, path safety, and scheduling robustness of multi-AGV systems in complex environments. Attached Figure Description

[0016] Figure 1 This is a block diagram of the present invention. Detailed Implementation

[0017] The subject matter described herein will now be discussed with reference to exemplary embodiments. It should be understood that these embodiments are discussed only to enable those skilled in the art to better understand and implement the subject matter described herein, and changes may be made to the function and arrangement of the elements discussed without departing from the scope of this specification. Various processes or components may be omitted, substituted, or added as needed in the examples. Furthermore, features described in some examples may be combined in other examples.

[0018] like Figure 1 As shown, the operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop includes: The data acquisition module is used to acquire point cloud data, perform voxel rasterization on the point cloud data to obtain a voxel set, and determine the point cloud density of each voxel in the voxel set. The voxel filtering module is used to calculate the global average point cloud density of the voxel set, and to filter low-density voxels in the voxel set based on the global average point cloud density; and to perform connected component analysis on the low-density voxels in the marked voxel set to form M low-density connected components. The cavity identification module is used to perform upper and lower cone fitting processing on the m-th low-density connected domain to determine the fitting error, and to label the cavity of the m-th low-density connected domain based on the fitting error. The point cloud filling module is used to determine the boundary average point cloud density of the cavity of the m-th low-density connected domain, and to perform point cloud filling processing on the cavity of the m-th low-density connected domain based on the boundary average point cloud density to form the updated cavity of the m-th low-density connected domain. The management operation module is used to determine the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain, and to trigger the management operation of the AGV system based on the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain; wherein, the management operation includes: rerouting management operation or deceleration management operation.

[0019] In one embodiment of the present invention, point cloud data is acquired, and the point cloud data is rasterized into voxels to obtain a voxel set. The point cloud density of each voxel in the voxel set is determined, including: The point cloud data is divided using resolution r to generate a set of voxels; The point cloud density corresponding to each voxel in the voxel set is the sum of the number of points within the voxel. The ratio of .

[0020] It should be noted that for the 3D point cloud data acquired by LiDAR, voxel rasterization is first performed to achieve spatial discretization and density quantization analysis. In practice, the voxelization resolution *r* needs to be reasonably set based on the geometric precision of the construction scenario (e.g., the installation tolerances for electromechanical pipelines are typically in the millimeter to centimeter range) and the point cloud resolution of the LiDAR sensor (for example, in building electromechanical engineering applications, *r* can be selected in the range of 5mm to 50mm, balancing computational efficiency and preservation of geometric details). The 3D space is divided into a set of cubic voxels with side length *r*, each voxel corresponding to a unique spatial coordinate interval. For each voxel, the number of point cloud data points contained within it is counted, and the number of point cloud data points is compared with... The point cloud density of the corresponding voxel is obtained by calculating the ratio.

[0021] By using regularized voxel meshes, discrete and disordered point cloud data is transformed into spatial units with well-defined topological relationships, providing an efficient computational foundation and significantly reducing the time complexity of neighborhood search. Point cloud density, as a core quantitative indicator, directly characterizes the density of point clouds in a spatial region, ensuring accurate screening and geometric analysis of cavity phenomena in subsequent steps.

[0022] In one embodiment of the present invention, the global average point cloud density of the voxel set is calculated to filter low-density voxels in the voxel set based on the global average point cloud density. Connectivity analysis is then performed on the low-density voxels in the labeled voxel set to form M low-density connected components, including: The global average point cloud density of the voxel set is calculated, and the product of the preset scaling factor and the global average point cloud density is used as the filtering threshold. Voxels with point cloud density less than the screening threshold in the voxel set are classified as low-density voxels. The 26 neighborhoods of low-density voxels in the voxel set are merged to form M low-density connected regions.

[0023] It should be noted that, to accurately identify the low-density spatial region corresponding to the cavity phenomenon, it is necessary to first quantify the global average point cloud density of the voxel set and construct continuous spatial units through connected component merging. Specifically, firstly, all voxels in the voxel set are traversed. Based on the global average point cloud density (obtained in the same way as the point cloud density of each voxel), and considering the characteristic that the point density inside the funnel-shaped hollow corresponding to the cavity phenomenon is significantly lower than that of the environment, a preset scaling factor is introduced and multiplied by the global average point cloud density to construct a low-density screening threshold. Subsequently, the voxel set is traversed to screen voxels whose point cloud density is less than the low-density screening threshold, and these are marked as low-density voxels. Since the funnel-shaped hollows are continuously distributed in three-dimensional space, a 26-neighborhood connected component analysis needs to be performed on the low-density voxels: starting with a single low-density voxel, a breadth-first search (BFS) is used to traverse its 26 neighboring voxels in three-dimensional space (covering the horizontal and vertical axes, the positive and negative directions of the vertical axis, and adjacent voxels in the diagonal direction), and the connected low-density voxels are merged into the same set, ultimately forming M low-density connected components.

[0024] In one embodiment of the present invention, the m-th low-density connected component undergoes upper and lower cone fitting to determine the fitting error, including: Extract the boundary point cloud set of the m-th low-density connected component; The line connecting the data acquisition module and the target suspended object's origin is used as the observation axis. The axial projection value of the line connecting each boundary point cloud in the boundary point cloud set to the target suspended object's origin is determined in the observation axis direction. Based on the axial projection value of each boundary point cloud, the boundary point cloud set is divided into: a lower region set, a transition region set, and an upper region set. The lower centroid point cloud, transition centroid point cloud, and upper centroid point cloud of the coverage area of ​​the lower region set, transition region set, and upper region set are determined respectively. With the transition centroid point cloud as the constraint center, a symmetry axis is fitted. The symmetry axis is obtained by minimizing the radial deviation distance between the lower centroid point cloud and the upper centroid point cloud and the symmetry axis. For the boundary point clouds in each set of the lower region set and the upper region set, the projection points of the boundary point clouds on the symmetry axis are determined, and the projection points, boundary point clouds, and transition centroid point clouds are connected in sequence to form a triangle. The included angle between the two sides adjacent to the transition centroid point cloud in the triangle is determined. If the included angle is within a preset angle range, the corresponding boundary point cloud is marked as a candidate boundary point cloud. N sectional planes are loaded with a symmetry axis. If the ratio of the number of candidate boundary point clouds on both sides of the nth sectional plane is not within the preset ratio range, the upper and lower limits of the preset angle range are scaled down proportionally to regenerate the candidate boundary point cloud until the ratio of the number of candidate boundary point clouds on both sides of the N sectional planes is within the preset ratio range.

[0025] It should be noted that, to construct the axial spatial analysis benchmark of the funnel-shaped hollow structure, the definition logic of the observation axis is first clarified: the observation axis is defined as the line connecting the data acquisition module (such as a LiDAR sensor, with its spatial coordinates marked O1) and the point where the target suspended object occurs (typically a weld point in a welding operation, with its spatial coordinates marked O2). A unit direction vector is constructed through vector normalization. , This represents the spatial vector pointing from the sensor origin O1 to the solder joint O2. for The modulus length; for the boundary point cloud set of the m-th low-density connected domain, calculate the point cloud... Vector of the line connecting to solder joint O2 The axial projection value is × The axial projection value quantifies the spatial distribution of the point cloud along the observation axis relative to the weld point O2 in a physical sense. That is, the smaller the axial projection value, the closer the point cloud is to the weld pool side (lower funnel region); the larger the axial projection value, the closer the point cloud is to the welding fume rising side (upper funnel region).

[0026] It should be noted that the axial projection values ​​of all point clouds in the boundary point cloud set are sorted in ascending order to generate a sequence. Based on the axial distribution characteristics of the funnel-shaped hollow structure (the lower funnel is close to the molten pool and has sparse point clouds; the transition zone connects the two funnels and has relatively dense point clouds; the upper funnel extends with welding fumes and has sparse point clouds), a dynamic quantile method is used to divide the region: the 30% quantile *tlow* and the 70% quantile *tup* of the sequence are calculated. Point clouds with axial projection values ​​< *tlow* are assigned to the lower region set (corresponding to the lower funnel near the molten pool region), point clouds with axial projection values ​​> *tup* are assigned to the upper region set (corresponding to the upper funnel far from the molten pool region), and the remaining point clouds are assigned to the transition region set (corresponding to the double funnel connection region).

[0027] It should be noted that the three sets of centroid point clouds of the m-th low-density connected region are denoted as follows: Lower center of mass point cloud: ; Transitional centroid point cloud: ; Upper center of mass ; The transition centroid point cloud is set as the constraint center; the axis of symmetry to be determined is represented by a vector. It indicates that, and satisfies .

[0028] For centroid point cloud Its radial deviation from the axis of symmetry Defined as: in, This represents the vector from the lower centroid point cloud to the transition centroid point cloud (constraint center); express The length of the projection along the axis of symmetry.

[0029] Similarly, we can obtain the center-of-mass cloud. Its radial deviation from the axis of symmetry .

[0030] Construct the objective function ; The axis of symmetry is obtained by solving the eigenvalues ​​based on the objective function.

[0031] It should be noted that, firstly, for any boundary point cloud, its projection points on the axis of symmetry are calculated. According to the principle of vector projection, the projection points... satisfy: .

[0032] Then, connect the projection points Point clouds Transition centroid point cloud. The included angle between two adjacent sides of the transition centroid point cloud in the triangle is calculated using the vector dot product formula. If the included angle is within a preset angle range (preferably 20 to 30 degrees, corresponding to a funnel-shaped hollow center). (the tolerance range of the outward angle), then mark the point cloud. For candidate boundary point clouds.

[0033] It should be noted that when loading N sectional planes with the axis of symmetry, the axis of symmetry for each sectional plane is a straight line within that sectional plane.

[0034] In one embodiment of the present invention, the m-th low-density connected component undergoes upper and lower cone fitting to determine the fitting error, and further includes: K-clustering is performed on the candidate boundary point cloud to form a lower funnel sub-cluster set and an upper funnel sub-cluster set. The lower funnel sub-cluster set and the upper funnel sub-cluster set are fitted with random sampling consensus algorithm to obtain the lower cone model and the upper cone model respectively. The lower cone model includes: the lower cone vertex, the lower axis unit vector, and the lower half-vertex angle; the upper cone model includes: the upper cone vertex, the upper axis unit vector, and the upper half-vertex angle. The fitting error is constructed based on the difference between the lower half-vertex angle and the upper half-vertex angle, the sum of the lower axis unit vector and the upper axis unit vector, and the vertical distance difference between the lower cone vertex and the upper cone vertex.

[0035] It should be noted that the K-means clustering algorithm is used to divide the candidate boundary point clouds into subclusters using the axial projection values, with the number of clusters set to 2. Point clouds with smaller axial projection values ​​are assigned to the lower funnel subcluster set, and point clouds with larger axial projection values ​​are assigned to the upper funnel subcluster set. For the lower and upper funnel subcluster sets, the RANSAC algorithm is used to fit conical models, resulting in the following models: Lower Cone Model: Output parameters are the lower cone vertex (spatial coordinates, closer to the molten pool), lower axial unit vector (aligned with the axis of symmetry, pointing towards the molten pool), and lower half-vertex angle (convergence angle of the lower funnel); Upper Cone Model: Output parameters are the upper cone vertex (spatial coordinates, farther from the molten pool), upper axial unit vector (aligned with the axis of symmetry, away from the molten pool), and upper half-vertex angle (outward angle of the upper funnel). The fitting error is obtained by summing the difference between the lower and upper half-vertex angles, the sum of the lower and upper axial unit vectors, and the vertical distance difference between the lower and upper cone vertices.

[0036] In one embodiment of the present invention, the cavity of the m-th low-density connected region is labeled based on the fitting error, comprising: Calculate the ratio of the fitting error to the preset cavity threshold; If the ratio is greater than 1, then the m-th low-density connected component is marked as a cavity; If the ratio is ≤1, no marking is performed.

[0037] It should be noted that a ratio greater than 1 indicates that the geometric deviation between the fitting result and the hollow funnel shape exceeds the theoretical tolerance.

[0038] In one embodiment of the present invention, the boundary average point cloud density of the cavity of the m-th low-density connected domain is determined, and point cloud filling processing is performed on the cavity of the m-th low-density connected domain according to the boundary average point cloud density to form an updated cavity of the m-th low-density connected domain, including: Identify the voxels in the m-th low-density connected region that are in contact with the lower cone model or the upper cone model, form a boundary voxel set, and calculate the boundary average point cloud density of the boundary voxel set; The horizontal reference height is obtained by taking the difference between the horizontal heights of the lower cone vertex and the upper cone vertex of the lower cone model or the upper cone model, and then taking half of the difference as the horizontal height of the lower cone vertex. Construct a horizontal target density function based on the horizontal reference height and the boundary average point cloud density; For the horizontal target density function, determine the target point cloud density of each voxel in the m-th low-density connected region; based on the point cloud density difference between the target point cloud density and the point cloud density of each voxel in the m-th low-density connected region, determine the filling quantity of the virtual point cloud of the corresponding voxel, and perform uniform filling processing based on the filling quantity of the virtual point cloud to form the updated cavity of the m-th low-density connected region.

[0039] It should be noted that the horizontal target density function is constructed based on the horizontal reference height and the average point cloud density at the boundary. ,in, Indicates the attenuation coefficient. Indicates the horizontal height of the point cloud. This represents the target density corresponding to the horizontal height z. The boundary average point cloud density is represented by the horizontal target density function, which makes the target density decrease symmetrically with height, satisfying the property of a funnel-shaped hollow center with sparse ends.

[0040] In one embodiment of the present invention, determining the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected region includes: Determine the number of virtual point clouds and the point cloud density in the vth voxel of the boundary voxel set, and calculate the voxel confidence and voxel occupancy probability of the vth voxel, as follows: in, This represents the voxel confidence level of the vth voxel. Let v represent the voxel occupancy probability, specifically the voxel occupancy probability of the vth voxel. This represents the confidence decay coefficient. This represents the number of virtual point clouds filled for the v-th v element. Represents the natural base. This represents the cumulative occupancy coefficient. This represents the point cloud density of the vth voxel.

[0041] It should be noted that the more virtual point cloud fills, the sparser the original point cloud, and the lower the reliability of the repaired data; as the virtual point cloud fills, the voxel confidence index decays. The total density of the real and virtual point clouds is mapped to occupancy probability: the higher the density, the lower the occupancy probability. The closer the value is to 0, the closer the occupancy probability is to 1 (indicating that the voxel is more likely to be occupied rather than free space), accurately characterizing the semi-occluded property of the repaired cavity (neither completely empty nor a physical obstacle). This integrates the occupancy model of real and virtual density, correcting the defect of traditional SLAM that misclassifies cavities as free space.

[0042] In one embodiment of the present invention, the management operation of the AGV system is triggered based on the voxel confidence and voxel occupancy probability of each voxel in the updated cavity of the m-th low-density connected domain, including: Determine the average voxel confidence of the boundary voxel set and average voxel occupancy probability ; Calculate the cavity risk score based on the average voxel confidence level and the average voxel occupancy probability: in, Indicates the cavity risk score; The cavity risk score is compared with a preset risk threshold. When the cavity risk score is greater than or equal to the preset risk threshold, the AGV system's management operation is triggered, as follows: If it is determined that the number of switching paths of the device to which the AGV system belongs is ≥2 at the trigger time, then the device to which the AGV system belongs will be controlled to perform a rerouting management operation. If the number of switching paths in the device to which the AGV system belongs at the trigger time is 1, then control the device to which the AGV system belongs to perform deceleration management operation; When no low-density connected domain in the voxel set is marked as a cavity, the management operation of the AGV system is not triggered.

[0043] It should be noted that, To achieve synergistic quantification of credibility defects and space occupancy risks, and to avoid the one-sidedness of a single indicator (such as areas with high confidence but high occupancy probability, or areas with low confidence but low occupancy probability, can all reasonably reflect risks through multiplication).

[0044] The embodiments of this example have been described above. However, this example is not limited to the specific implementation methods described above. The specific implementation methods described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms based on the guidance of this example, and all of them are within the protection scope of this example.

Claims

1. An operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop, characterized in that, include: The data acquisition module is used to acquire point cloud data, perform voxel rasterization on the point cloud data to obtain a voxel set, and determine the point cloud density of each voxel in the voxel set. The voxel filtering module is used to calculate the global average point cloud density of the voxel set, and to filter low-density voxels in the voxel set based on the global average point cloud density; and to perform connected component analysis on the low-density voxels in the marked voxel set to form M low-density connected components. The cavity identification module is used to perform upper and lower cone fitting processing on the m-th low-density connected domain to determine the fitting error, and to label the cavity of the m-th low-density connected domain based on the fitting error. The point cloud filling module is used to determine the boundary average point cloud density of the cavity of the m-th low-density connected domain, and to perform point cloud filling processing on the cavity of the m-th low-density connected domain based on the boundary average point cloud density to form the updated cavity of the m-th low-density connected domain. The management operation module is used to determine the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain, and to trigger the management operation of the AGV system based on the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected domain; wherein, the management operation includes: rerouting management operation or deceleration management operation.

2. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 1, characterized in that, Collect point cloud data and perform voxel rasterization to obtain a voxel set. Determine the point cloud density of each voxel in the voxel set, including: The point cloud data is divided using resolution r to generate a set of voxels; The point cloud density corresponding to each voxel in the voxel set is the sum of the number of points within the voxel. The ratio of .

3. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 2, characterized in that, Calculate the global average point cloud density of the voxel set, and use this density to filter low-density voxels. Perform connected component analysis on the low-density voxels in the labeled voxel set to form M low-density connected components, including: The global average point cloud density of the voxel set is calculated, and the product of the preset scaling factor and the global average point cloud density is used as the filtering threshold. Voxels with point cloud density less than the screening threshold in the voxel set are classified as low-density voxels. The 26 neighborhoods of low-density voxels in the voxel set are merged to form M low-density connected regions.

4. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 3, characterized in that, The m-th low-density connected component is subjected to upper and lower cone fitting to determine the fitting error, including: Extract the boundary point cloud set of the m-th low-density connected component; The line connecting the data acquisition module and the target suspended object's origin is used as the observation axis. The axial projection value of the line connecting each boundary point cloud in the boundary point cloud set to the target suspended object's origin is determined in the observation axis direction. Based on the axial projection value of each boundary point cloud, the boundary point cloud set is divided into: a lower region set, a transition region set, and an upper region set. The lower centroid point cloud, transition centroid point cloud, and upper centroid point cloud of the coverage area of ​​the lower region set, transition region set, and upper region set are determined respectively. With the transition centroid point cloud as the constraint center, a symmetry axis is fitted. The symmetry axis is obtained by minimizing the radial deviation distance between the lower centroid point cloud and the upper centroid point cloud and the symmetry axis. For the boundary point clouds in each set of the lower region set and the upper region set, the projection points of the boundary point clouds on the symmetry axis are determined, and the projection points, boundary point clouds, and transition centroid point clouds are connected in sequence to form a triangle. The included angle between the two sides adjacent to the transition centroid point cloud in the triangle is determined. If the included angle is within a preset angle range, the corresponding boundary point cloud is marked as a candidate boundary point cloud. N sectional planes are loaded with a symmetry axis. If the ratio of the number of candidate boundary point clouds on both sides of the nth sectional plane is not within the preset ratio range, the upper and lower limits of the preset angle range are scaled down proportionally to regenerate the candidate boundary point cloud until the ratio of the number of candidate boundary point clouds on both sides of the N sectional planes is within the preset ratio range.

5. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 4, characterized in that, The m-th low-density connected component undergoes upper and lower cone fitting to determine the fitting error, which also includes: K-clustering is performed on the candidate boundary point cloud to form a lower funnel sub-cluster set and an upper funnel sub-cluster set. The lower funnel sub-cluster set and the upper funnel sub-cluster set are fitted with random sampling consensus algorithm to obtain the lower cone model and the upper cone model respectively. The lower cone model includes: the lower cone vertex, the lower axis unit vector, and the lower half-vertex angle; the upper cone model includes: the upper cone vertex, the upper axis unit vector, and the upper half-vertex angle. The fitting error is constructed based on the difference between the lower half-vertex angle and the upper half-vertex angle, the sum of the lower axis unit vector and the upper axis unit vector, and the vertical distance difference between the lower cone vertex and the upper cone vertex.

6. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 5, characterized in that, The cavity of the m-th low-density connected region is labeled based on the fitting error, including: Calculate the ratio of the fitting error to the preset cavity threshold; If the ratio is greater than 1, then the m-th low-density connected component is marked as a cavity; If the ratio is ≤1, no marking is performed.

7. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 6, characterized in that, Determine the boundary average point cloud density of the cavity in the m-th low-density connected domain, and perform point cloud filling processing on the cavity in the m-th low-density connected domain based on the boundary average point cloud density to form the updated cavity in the m-th low-density connected domain, including: Identify the voxels in the m-th low-density connected region that are in contact with the lower cone model or the upper cone model, form a boundary voxel set, and calculate the boundary average point cloud density of the boundary voxel set; The horizontal reference height is obtained by taking the difference between the horizontal heights of the lower cone vertex and the upper cone vertex of the lower cone model or the upper cone model, and then taking half of the difference as the horizontal height of the lower cone vertex. Construct a horizontal target density function based on the horizontal reference height and the boundary average point cloud density; For the horizontal target density function, determine the target point cloud density of each voxel in the m-th low-density connected region; based on the point cloud density difference between the target point cloud density and the point cloud density of each voxel in the m-th low-density connected region, determine the filling quantity of the virtual point cloud of the corresponding voxel, and perform uniform filling processing based on the filling quantity of the virtual point cloud to form the updated cavity of the m-th low-density connected region.

8. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 7, characterized in that, Determine the voxel confidence and voxel occupancy probability of each voxel in the update cavity of the m-th low-density connected component, including: Determine the number of virtual point clouds and the point cloud density in the vth voxel of the boundary voxel set, and calculate the voxel confidence and voxel occupancy probability of the vth voxel, as follows: in, This represents the voxel confidence level of the vth voxel. Let v represent the voxel occupancy probability, specifically the voxel occupancy probability of the vth voxel. This represents the confidence decay coefficient. This represents the number of virtual point clouds filled for the v-th v element. Represents the natural base. This represents the cumulative occupancy coefficient. This represents the point cloud density of the vth voxel.

9. The operation monitoring system based on multiple AGV systems in a flexible manufacturing workshop according to claim 8, characterized in that, The AGV system's management tasks are triggered based on the voxel confidence and voxel occupancy probability of each voxel in the updated cavity of the m-th low-density connected region, including: Determine the average voxel confidence of the boundary voxel set and average voxel occupancy probability ; Calculate the cavity risk score based on the average voxel confidence level and the average voxel occupancy probability: in, Indicates the cavity risk score; The cavity risk score is compared with a preset risk threshold. When the cavity risk score is greater than or equal to the preset risk threshold, the AGV system's management operation is triggered, as follows: If it is determined that the number of switching paths of the device to which the AGV system belongs is ≥2 at the trigger time, then the device to which the AGV system belongs will be controlled to perform a rerouting management operation. If the number of switching paths in the device to which the AGV system belongs at the trigger time is 1, then control the device to which the AGV system belongs to perform deceleration management operation; When no low-density connected domain in the voxel set is marked as a cavity, the management operation of the AGV system is not triggered.

Images